 Combining airborne imagery with computer vision can generate individual tree data at large scales, but biased predictions occur when only common species are used to train the model. A targeted sampling workflow and rare species sampling extend classification models to include rare species, resulting in improved rare species classification and landscape species maps of individual crowns for over 670,000 individual trees. These maps provide estimates of canopy tree diversity within a neon site and highlight the importance of capturing species diversity in training data. This article was authored by Ben G. Weinstein, Sergio Marconi, Sarah J. Graves and others.